SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.11246 · ICML · 2026

Support-Proximity Augmented Diffusion Estimation for Offline Black-Box Optimization

Ye Yuan, Can Chen, Xue Liu, Bowei He, Haolun Wu, Linfeng Du, Zipeng Sun, Yonghan Yang

arXiv · PDF · Open in the Atlas

Code that ran

15 of 20 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 6 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 5 failures of our instrument; 5 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2605.11246.json · badge: /api/ran/2605.11246.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
HarryYoung2018/spade community
Papers with Code archive
15 ran · 5 no recorded run
FunctionStatusWhere it lives
CondEpsNet Ran
ran on a synthesized input
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("97a9f68b697701d9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math
import torch.nn as nn
import torch.nn.functional as F

class TimeEmbedding(nn.Module):

    def __init__(self, dim: int):
        super().__init__()
        self.dim = int(dim)
        self.lin1 = nn.Linear(self.dim, self.dim)
        self.lin2 = nn.Linear(self.dim, self.dim)

    def forward(self, t: torch.Tensor) -> torch.Tensor:
        if t.dim() == 1:
            t = t[:, None]
        half = self.dim // 2
        freqs = torch.exp(torch.linspace(0, math.log(10000), steps=half, device=t.device))
        args = t * freqs
        emb = torch.cat([torch.sin(args), torch.cos(args)], dim=-1)
        emb = F.silu(self.lin1(emb))
        return self.lin2(emb)

class CondEpsNet(nn.Module):

    def __init__(self, x_dim: int, hidden: int, t_dim: int):
        super().__init__()
        self.t_embed = TimeEmbedding(t_dim)
        self.x_enc = nn.Sequential(nn.Linear(x_dim, hidden), nn.SiLU(), nn.Linear(hidden, hidden), nn.SiLU())
        self.y_in = nn.Linear(1, hidden)
        self.fc = nn.Sequential(nn.Linear(hidden + hidden + t_dim, hidden), nn.SiLU(), nn.Linear(hidden, hidden), nn.SiLU(), nn.Linear(hidden, 1))

    def forward(self, y_t: torch.Tensor, t_cont: torch.Tensor, x: torch.Tensor) -> torch.Tensor:
        t_emb = self.t_embed(t_cont.view(-1))
        xh = self.x_enc(x)
        yh = self.y_in(y_t)
        h = torch.cat([xh, yh, t_emb], dim=-1)
        return self.fc(h)


INIT_ARGS = {
    'x_dim': 4,
    'hidden': 8,
    't_dim': 6,
}

ARGS = {
    'y_t': torch.randn(2, 1),
    't_cont': torch.randn(2),
    'x': torch.randn(2, 4),
}

if __name__ == '__main__':
    model = CondEpsNet(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 1) — shape (2, 1) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
KnnDensityHelper Ran
ran on a synthesized input
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("9f70f55965dff86a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned ndarray, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from sklearn.neighbors import NearestNeighbors

def _as_2d(arr: np.ndarray) -> np.ndarray:
    arr = np.asarray(arr, dtype=np.float32)
    if arr.ndim == 1:
        return arr[None, :]
    return arr

class KnnDensityHelper:

    def __init__(self, x_train: np.ndarray, max_k: int, *, transform=None):
        self.transform = transform
        x_fit = transform(x_train) if transform is not None else x_train
        x_fit = _as_2d(x_fit)
        self.x = x_fit
        self.dim = x_fit.shape[1]
        self.n = x_fit.shape[0]
        self.nn = NearestNeighbors(n_neighbors=max_k).fit(x_fit)
        self.max_k = int(max_k)

    def _apply(self, x_query: np.ndarray) -> np.ndarray:
        return self.transform(x_query) if self.transform is not None else x_query

    def log_rk(self, x_query: np.ndarray, k: int) -> np.ndarray:
        x_query = _as_2d(x_query)
        x_query = self._apply(x_query)
        k = int(k)
        if k <= 0 or k > self.max_k:
            raise ValueError(f'Invalid k={k} for kNN helper (max_k={self.max_k}).')
        dists = self.nn.kneighbors(x_query, n_neighbors=k, return_distance=True)[0]
        r_k = dists[:, -1] + 1e-08
        return np.log(r_k)


INIT_ARGS = {
    'x_train': np.random.randn(8, 4).astype(np.float32),
    'max_k': 5,
    'transform': None,
}

ARGS = {
    'x_query': np.random.randn(2, 4).astype(np.float32),
    'k': 3,
}

METHOD = 'log_rk'

if __name__ == '__main__':
    helper = KnnDensityHelper(**INIT_ARGS)
    result = helper.log_rk(**ARGS)
    print(f"Result shape: {result.shape}")
    print(f"Result: {result}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ScalarDiffusionSurrogate Ran
ran on a synthesized input
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("681458c13d0167a2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math
import torch.nn as nn
import torch.nn.functional as F

# Define INIT_ARGS for ScalarDiffusionSurrogate.__init__
INIT_ARGS = {
    'x_dim': 4,
    'steps': 10,
    'hidden': 8,
    't_dim': 6,
    'beta_start': 0.0001,
    'beta_end': 0.02,
    'device': torch.device('cpu'),
}

# Define ARGS for the computation method (loss)
ARGS = {
    'x': torch.randn(2, 4),
    'y': torch.randn(2, 1),
}

# Define the method to call
METHOD = 'loss'
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TimeEmbedding Ran
ran on a synthesized input
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("4cf430efa36c0fb8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math
import torch.nn as nn
import torch.nn.functional as F

class TimeEmbedding(nn.Module):

    def __init__(self, dim: int):
        super().__init__()
        self.dim = int(dim)
        self.lin1 = nn.Linear(self.dim, self.dim)
        self.lin2 = nn.Linear(self.dim, self.dim)

    def forward(self, t: torch.Tensor) -> torch.Tensor:
        if t.dim() == 1:
            t = t[:, None]
        half = self.dim // 2
        freqs = torch.exp(torch.linspace(0, math.log(10000), steps=half, device=t.device))
        args = t * freqs
        emb = torch.cat([torch.sin(args), torch.cos(args)], dim=-1)
        emb = F.silu(self.lin1(emb))
        return self.lin2(emb)


INIT_ARGS = {
    'dim': 4
}

ARGS = {
    't': torch.randn(2)
}

if __name__ == '__main__':
    model = TimeEmbedding(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
_as_2d Ran
ran, honoured the contract we drafted
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("80a38ab4a9fb9601")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ndarray, ndim 2, dtype family float, shape (1, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Test case 1: 1D array (should be converted to 2D)
arr_1d = np.array([1.0, 2.0, 3.0, 4.0], dtype=np.float32)

# Test case 2: 2D array (should remain 2D)
arr_2d = np.array([[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]], dtype=np.float32)

# Use the 1D array as the primary test case
ARGS = {
    "arr": arr_1d
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (1, 4) — shape (1, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
_ensure_2d Ran
ran, honoured the contract we drafted
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("ab5cef763a973407")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ndarray, ndim 2, dtype family int, shape (5, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Test case 1: 1D array that should be converted to 2D
arr_1d = np.array([1, 2, 3, 4, 5])

# Test case 2: 2D array that should remain unchanged
arr_2d = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])

# Test case 3: List that will be converted to array
arr_list = [10, 20, 30, 40]

# Test case 4: Nested list (2D)
arr_nested = [[1, 2], [3, 4], [5, 6]]

ARGS = {
    "arr": arr_1d
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family int, shape (5, 1) — shape (5, 1) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
acquisition_from_samples Ran
ran; our contract draft was wrong, not the code
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("4db6557dac754a47")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 1, dtype family float, shape (4,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create sample data: 2 samples, 4 dimensions
samples = torch.randn(2, 4)

# Set y_best to a reasonable value
y_best = 0.5

# Set acquisition function kind
kind = 'ei'

# Set ucb_beta parameter
ucb_beta = 2.0

# Set ei_xi parameter (expected improvement exploration parameter)
ei_xi = 0.1

# Set mvr_lambda parameter (mean-variance regularization)
mvr_lambda = 0.5

ARGS = {
    'samples': samples,
    'y_best': y_best,
    'kind': kind,
    'ucb_beta': ucb_beta,
    'ei_xi': ei_xi,
    'mvr_lambda': mvr_lambda,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (4,) — shape (4,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
apply_support_transform Ran
ran; our contract draft was wrong, not the code
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("855e6bffce251fe6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create sample tensors
# samples: shape (batch_size, feature_dim) - used for mean/std computation
samples = torch.randn(2, 4)

# log_rk: shape (n_features,) or broadcastable - used for adjustment calculations
log_rk = torch.randn(4)

# Scalar parameters
tau_a = 0.5
sigma_a0 = 0.1
sigma_a1 = 0.2

ARGS = {
    "samples": samples,
    "log_rk": log_rk,
    "tau_a": tau_a,
    "sigma_a0": sigma_a0,
    "sigma_a1": sigma_a1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
build_knn_helper Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("71870850a265ad4e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned KnnDensityHelper. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from typing import Callable, Sequence

def dummy_transform(x: np.ndarray) -> np.ndarray:
    """Simple transform that scales data by 2"""
    return x * 2.0

# Create small training data: 5 samples, 3 features
x_train = np.random.randn(5, 3).astype(np.float32)

# Define k_list with valid values
k_list = [1, 2, 3]

# Define transform as None or a callable
transform = None

ARGS = {
    "x_train": x_train,
    "k_list": k_list,
    "transform": transform,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type KnnDensityHelper — a KnnDensityHelper output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a KnnDensityHelper output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
clip_and_renorm Ran
ran, honoured the contract we drafted
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("d1fe61e95e0ab9cf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ndarray, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Set up small numpy arrays for the function parameters
# x_norm: normalized input array (batch_size=2, features=4)
x_norm = np.array([
    [0.5, -0.3, 1.2, -0.8],
    [0.1, 0.4, -0.6, 0.9]
], dtype=np.float32)

# bounds: shape (4, 2) - lower and upper bounds for each feature
bounds = np.array([
    [-1.0, 1.0],
    [-2.0, 2.0],
    [-0.5, 0.5],
    [-1.5, 1.5]
], dtype=np.float32)

# x_mean: mean for each feature (4,)
x_mean = np.array([0.1, 0.2, -0.1, 0.05], dtype=np.float32)

# x_std: standard deviation for each feature (4,)
x_std = np.array([0.5, 0.8, 0.3, 0.6], dtype=np.float32)

ARGS = {
    "x_norm": x_norm,
    "bounds": bounds,
    "x_mean": x_mean,
    "x_std": x_std
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
clip_to_bounds Ran
ran, honoured the contract we drafted
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("d01936d34e41a35c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ndarray, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small input array with shape (2, 4)
x = np.array([
    [-1.0, 0.5, 2.0, 10.0],
    [0.0, 1.5, 3.0, 5.0]
], dtype=np.float32)

# Create bounds array with shape (4, 2) where each row is [lower, upper]
bounds = np.array([
    [-0.5, 0.5],    # bounds for dimension 0
    [0.0, 1.0],     # bounds for dimension 1
    [1.0, 2.5],     # bounds for dimension 2
    [0.0, 4.0]      # bounds for dimension 3
], dtype=np.float32)

ARGS = {
    "x": x,
    "bounds": bounds
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
evolutionary_optimize Ran
ran; our contract draft was wrong, not the code
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("36bd257d56f94f61")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from typing import Callable

# Set random seed for reproducibility
np.random.seed(0)
rng = np.random.default_rng(0)

# Define dimensions: feature_dim=4, sample_dim=2
feature_dim = 4
n_samples = 2

# Create training data
x_train = np.random.randn(n_samples, feature_dim).astype(np.float32)
y_train = np.random.randn(n_samples).astype(np.float32)

# Create bounds for each feature (lower and upper bounds)
bounds = np.array([
    [-1.0, 1.0],
    [-2.0, 2.0],
    [-0.5, 0.5],
    [-3.0, 3.0]
], dtype=np.float32)

# Create normalization parameters
x_mean = np.mean(x_train, axis=0).astype(np.float32)
x_std = np.std(x_train, axis=0).astype(np.float32) + 1e-8  # avoid division by zero

# Create a simple score function that takes a population and returns scores
def score_fn(pop: np.ndarray) -> np.ndarray:
    """Simple scoring function: sum of features"""
    return np.sum(pop, axis=1).astype(np.float32)

ARGS = {
    'score_fn': score_fn,
    'x_train': x_train,
    'y_train': y_train,
    'bounds': bounds,
    'x_mean': x_mean,
    'x_std': x_std,
    'rng': rng,
    'pop_size': 4,
    'elite_size': 2,
    'gens': 2,
    'mut_sigma_init': 0.1,
    'mut_sigma_min': 0.01,
    'crossover_rate': 0.5,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_beta_schedule Ran
ran, honoured the contract we drafted
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("146db3f324ec3821")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 1, dtype family float, shape (100,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "steps": 100,
    "beta_start": 0.0001,
    "beta_end": 0.02,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (100,) — shape (100,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
z_score Ran
ran; our contract draft was wrong, not the code
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("80a3da03a6fc9ab7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small numpy array with distinct dimensions
# Using shape (2, 4) - batch size 2, feature dimension 4
arr = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0]
], dtype=np.float32)

ARGS = {
    "arr": arr
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
z_score_denorm Ran
ran, honoured the contract we drafted
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("f39bb60e2d4fae07")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ndarray, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create small test arrays for z-score denormalization
# The function takes normalized data and reverses the z-score normalization
# by multiplying by std and adding mean

arr = np.array([[1.0, 2.0, 3.0, 4.0],
                [5.0, 6.0, 7.0, 8.0]], dtype=np.float32)

mean = np.array([10.0, 20.0, 30.0, 40.0], dtype=np.float32)

std = np.array([2.0, 3.0, 4.0, 5.0], dtype=np.float32)

ARGS = {
    "arr": arr,
    "mean": mean,
    "std": std
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Dataset Not yet run
unverified (harvested, no recorded run)
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("1fe070d1ecb71daf")
How it ran

no invocation recorded.

OptimizationResult Not yet run
unverified (harvested, no recorded run)
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("42cd3d3a49f9da3c")
How it ran

no invocation recorded.

SpadeConfig Not yet run
unverified (harvested, no recorded run)
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("4d8039a217216ef4")
How it ran

no invocation recorded.

mc_samples Not yet run
unverified (harvested, no recorded run)
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("47bdbaeb1cc4a7f4")
How it ran

no invocation recorded.

optimize_spade Not yet run
unverified (harvested, no recorded run)
HarryYoung2018/spade/spade/optimize.py
code served (permissive licence) · get_code("3ddf62b9dc3c9a75")
How it ran

no invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 20 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 20 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Offline black-box optimization aims to discover novel designs with high property scores using only a static dataset, a task fundamentally challenged by the out-of-distribution (OOD) extrapolation problem. Existing approaches typically bifurcate into inverse methods, which struggle with the ill-posed nature of mapping scores to designs, and forward methods, which often lack the distributional expressivity to quantify uncertainty effectively. In this work, we propose SPADE (Support-Proximity Augmented Diffusion Estimation), a novel framework that reimagines forward surrogate modeling through the lens of conditional generative modeling. SPADE models the forward likelihood p(y|x) using a diffusion model, but with two critical enhancements to tailor it for optimization: (1) a Calibrated Diffusion Estimation module that enforces global consistency in statistical moments and pairwise rankings, and (2) a Support-Proximity Regularization mechanism that implicitly internalizes the data manifold constraint p(x) via kNN-based density estimation. Theoretically, we prove that our regularization is first-order equivalent to maximizing a Bayesian posterior with a valid design prior. Empirically, SPADE achieves state-of-the-art performance across Design-Bench tasks and an LLM data mixture optimization benchmark. Our code is available at https://github.com/HarryYoung2018/spade. * Equal contribution with random order: Ye designs algorithm/drafts paper; Yonghan conducts experiments. †This work is done during Ye's visit at MBZUAI. ‡Work done independent of the author's position at Amazon AGI.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2605.11246")
get_code_for_paper("2605.11246")
have("2605.11246")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2605.11246.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2605.11246.svg)](https://syntology.ai/paper/2605.11246)

Connect an agent — have() is free.